What Is a Spectrogram?

That third axis is the whole point. A waveform tells you how loud the signal is overall; a spectrogram tells you which frequencies make up that loudness and how they move. Two completely different sounds can share an identical waveform shape and look totally different on a spectrogram.

How to read a spectrogram: what the shapes mean

Sounds leave recognisable fingerprints. Learn these five and you can diagnose a mix by eye:

The skill is pattern recognition, not maths. After a few sessions you'll glance at a vocal and immediately see the sibilant flares around 6 to 9 kHz, the breath noise between phrases, and whether the low end is clean or clouded with proximity boom.

Spectrogram vs spectrum analyzer: what's the difference?

They're cousins, and people mix up the names. A spectrum analyzer (the bouncing curve in your EQ) shows the frequency balance at a single instant, level on the Y axis, frequency on the X. A spectrogram rotates frequency onto the Y axis and uses the X axis for time, so it shows how that balance evolves. The analyzer answers 'what's loud right now?'; the spectrogram answers 'what happened across the whole song?'. You want both, for different jobs: the analyzer for live tone-shaping, the spectrogram for hunting problems that only show up over time.

What engineers actually use a spectrogram for

It's not eye candy. A spectrogram solves specific, recurring problems that are hard to catch any other way:

• Finding resonances. A ringing frequency shows up as a bright horizontal line you can park an EQ notch on, instead of sweeping a boost around hoping to hear it.
• Spotting masking. When two instruments fight in the same band, you see their energy overlapping in the same horizontal zone. That's your cue to carve a pocket with EQ.
• Hunting noise and hum. Ground-loop hum, air-conditioner rumble and digital glitches all have unmistakable signatures that hide under the music until you look.
• Checking the top end. A hard shelf above 16 to 20 kHz is the tell-tale fingerprint of a file that was once an MP3, which matters when you're handed 'masters' to work from.
• De-essing and restoration. Sibilant flares and mouth clicks are obvious on a spectrogram, so you can target them surgically instead of by ear alone.

That last point about the MP3 fingerprint is worth dwelling on. If someone sends you a 'WAV' and the spectrogram shows a brick-wall cliff at 16 kHz, the audio was lossy at some point in its life and re-wrapped, no amount of processing brings back what the encoder threw away. It's a strong argument for insisting on the original lossless file. When you hand off your own work, sending the real export through a delivery Room keeps that high end intact instead of letting a chat app re-encode it down to a cliff.

Reading loudness off a spectrogram (and where it stops)

Brightness tells you relative level frequency-by-frequency, which is perfect for balance decisions. What it doesn't give you is a calibrated loudness number for delivery, the integrated LUFS your streaming platform cares about. For that you need a loudness meter, not a colour scale. Use the spectrogram to fix what's loud, and a LUFS meter to decide how loud the finished file should be.

The two views are complementary. A spectrogram might reveal a harsh 3 kHz buildup that's making the track feel louder and more fatiguing than the meter suggests; tame it, and you can often push the overall level higher before it gets harsh. Problems you can see are problems you can fix with intent rather than guesswork, the same reason engineers also keep an eye on transients when shaping punch.